Awaken Mode

Awaken Mode turns your AI from a passive assistant into a proactive operator. You define “watches” — trigger conditions paired with tasks — and the Agent automatically executes them and delivers results. Think: automatic server health reports every morning, instant alerts when error logs spike, or weekly summary generation.

What Is Awaken Mode?

In normal mode, the AI only works when you send a message. Awaken Mode gives it the ability to sense events and act independently:

Normal Mode	Awaken Mode
You ask, AI answers	AI monitors and notifies you
You manually check systems	Automatic scheduled inspections
Post-incident troubleshooting	Real-time anomaly detection
Repeat instructions every time	Configure once, runs continuously

Analogy: Normal mode is like an assistant waiting for instructions. Awaken Mode is like having an on-call engineer watching your systems 24/7 — you set the rules, it does the watching.

Core Concepts

Watch

A “watch” is a monitoring configuration that defines three things:

When to act (Trigger) — schedule, file change, new email, etc.
What to do (Task prompt) — a natural-language description of what the Agent should do
Where to send results (Output channel) — IM push, in-app chat, system notification, or silent log

Triggers

Triggers determine when a watch fires:

Trigger Type	Description	Use Cases
Cron	Cron expression schedule	Every day at 9 AM, every Monday at 10 AM
Interval	Fixed time intervals	Every 30 minutes, every 2 hours
File change	File/directory created, modified, or deleted	Log file updated, config file changed
New email	New email arrives	Alert emails, important client messages
Calendar event	N minutes before an event starts	Meeting reminders, deadline alerts
Webhook	External HTTP POST callback	CI/CD completion, monitoring alerts
Heartbeat	Agent periodic health check	System health detection
Manual	Triggered only from the panel	On-demand one-off checks

Sensors

Sensors are the low-level event detectors. They run continuously but consume no AI resources — they only trigger an Agent execution when an event is detected.

Built-in sensors:

Heartbeat Sensor: Periodic system status checks
File Watch Sensor: Monitors file system changes
Calendar Sensor: Detects upcoming calendar events
Email Sensor: Detects new incoming emails
App Lifecycle Sensor: Detects app startup/shutdown and milestone events

Output Channels

Where watch results are delivered:

Channel	Description	Best For
IM push	Send to DingTalk/Feishu/WeCom/Slack/Telegram	Urgent notifications, mobile access
In-app chat	Display in SailFish’s chat area	Non-urgent but worth reviewing
System notification	OS-level notification	Brief reminders
Silent log	Log only, no notification	Record keeping, later analysis

Creating Watches

Method 1: Talk to the Agent (Recommended)

Describe what you want in natural language and the Agent will create the watch automatically:

Check server status every morning at 9 AM and notify me on DingTalk.

Monitor /var/log/nginx/error.log — if new 500 errors appear, alert me on Feishu.

Every Friday at 4 PM, generate a weekly work summary and send it to WeCom.

The Agent understands your intent and configures the trigger, task description, and output channel automatically.

Method 2: Awaken Panel

Click the Awaken panel icon in the left sidebar (lightbulb shape)
Click “New Watch”
Choose a built-in template for a quick start, or configure from scratch:
- Select trigger type and parameters
- Write the task description (what the AI should do)
- Choose the output channel

Built-in Templates

SailFish provides templates for common scenarios — select and create in one click:

Template	Trigger	What It Does
Daily Morning Briefing	Daily schedule	Pushes weather, calendar, email summary, and news
Calendar Event Reminder	Before event	Smart reminders with context information
Email Smart Classification	New email	Auto-classifies; important emails notified instantly
Log File Monitoring	File change	Analyzes logs; alerts on errors
Server Health Check	Scheduled	Checks CPU, memory, disk, and service status
File Change Notification	File change	Monitors directory changes and notifies

The Awaken Panel

Click the Awaken icon in the sidebar to open the panel, where you can:

View all watches — see status (running / paused / error)
Pause / Resume — temporarily disable or re-enable a watch
Edit — modify trigger conditions or task description
Delete — remove watches you no longer need
Execution history — see each trigger time, result, and AI analysis
Sensor events — raw events detected by the underlying sensors

Advanced Features

Pre-check Mechanism

Before executing, the AI first evaluates whether it should run. For example, a log monitoring watch triggers, but the AI checks and finds no new errors — so it skips execution, avoiding unnecessary notifications.

This is called Agency of Omission — the AI has the authority to decide “this time, nothing needs to be done.”

Stateful Workflows

Watches remember their state across executions, enabling comparative analysis:

Server health check: “CPU was 50% last time, now it’s 85% — significant increase”
Log monitoring: “Last checked up to line 1000, starting from line 1001 this time”
Watches can also share state with each other for coordinated workflows

Isolated Agent

Awaken Mode uses a dedicated Agent instance (__watch__), completely isolated from your regular conversations:

Won’t interfere with your ongoing chat sessions
Has its own context and memory
All watches execute sequentially — no concurrency conflicts

Tips

Start with templates: Pick a template and customize — much faster than building from scratch
Pair with IM: Set output to an IM channel so you get notifications on your phone
Don’t create too many watches: More watches = more frequent AI execution; keep it manageable
Leverage pre-check: Let the AI decide whether to act, reducing notification noise

For practical configuration examples, see Watch Recipes.